Upgrading sklearn to 1.0.2

README.rst

@@ -174,7 +174,7 @@ If you want to run local integration tests, then use:
 
     # Required arguments for integration tests are found in test/conftest.py
 
-    pytest test/integration/local --docker-base-name <your_docker_image> \
+    pytest test/integration --docker-base-name <your_docker_image> \
                       --tag <your_docker_image_tag> \
                       --py-version <2_or_3> \
                       --framework-version <Scikit-learn_version>
@@ -183,8 +183,8 @@ If you want to run local integration tests, then use:
 .. parsed-literal::
 
   # Example
-  pytest test/integration/local --docker-base-name preprod-sklearn ``\``
-                                --tag 1.0 ``\``
+  pytest test/integration --docker-base-name preprod-sklearn ``\``
+                                --tag 1.0-1-cpu-py3 ``\``
                                 --py-version 3 ``\``
                                 --framework-version |FRAMEWORK_VERSION|
 
@@ -242,4 +242,4 @@ SageMaker Scikit-learn Container is licensed under the Apache 2.0 License. It is
 .com, Inc. or its affiliates. All Rights Reserved. The license is available at:
 http://aws.amazon.com/apache2.0/
 
-.. |FRAMEWORK_VERSION| replace:: 0.23-1
+.. |FRAMEWORK_VERSION| replace:: 1.0-1

ci/buildspec-extension.yml

@@ -2,7 +2,7 @@ version: 0.2
 
 env:
   variables:
-    SKLEARN_FRAMEWORK_VERSION: "0.23-1"
+    SKLEARN_FRAMEWORK_VERSION: "1.0-1"
     EXTENSION_FRAMEWORK_VERSION: "2.5-1"
 
 phases:

ci/buildspec.yml

@@ -2,17 +2,26 @@ version: 0.2
 
 env:
   variables:
-    FRAMEWORK_VERSION: "0.23-1"
+    FRAMEWORK_VERSION: "1.0-1"
 
 phases:
   install:
     runtime-versions:
       docker: 18
-      python: 3.7
+      python: 3.9
   pre_build:
     commands:
     - echo Logging in to Amazon ECR...
     - $(aws ecr get-login --no-include-email --region $AWS_DEFAULT_REGION)
+    - echo Installing dependencies...
+    - curl -LO http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
+    - bash Miniconda3-latest-Linux-x86_64.sh -bfp /miniconda3
+    - export PATH=/miniconda3/bin:${PATH}
+    - conda install python=3.7
+    - conda update -y conda
+    - conda install -y 'requests<2.21'  # sagemaker-python-sdk requires requests<2.21
+    - python3 -m pip install pip==20.1  # The new pip denpendency resolver in 20.2+ can't resolve 1.0-1 and 0.90 dependencies
+    - python3 -m pip install .[test]
   build:
     commands:
     - echo Build started on `date`
@@ -22,6 +31,13 @@ phases:
     - pip install wheel setuptools
     - python setup.py bdist_wheel
     - docker build -t preprod-sklearn:$FRAMEWORK_VERSION-cpu-py3 -f docker/$FRAMEWORK_VERSION/final/Dockerfile.cpu .
+    - echo Running tox...
+    - printf "FROM preprod-sklearn:$FRAMEWORK_VERSION-cpu-py3\nADD . /app\nWORKDIR /app\nRUN python3 -m pip install .[test]" > Dockerfile.test
+    - docker build -t test-sklearn -f Dockerfile.test .
+    - docker run --rm -t test-sklearn sh -c 'pytest --cov=sagemaker_sklearn_container --cov-fail-under=60 test/unit'
+    - docker run --rm -t test-sklearn sh -c 'flake8 setup.py src test'
+    - echo Running container tests...
+    - pytest test/integration --docker-base-name preprod-sklearn --tag $FRAMEWORK_VERSION-cpu-py3 --py-version 3 --framework-version $FRAMEWORK_VERSION
     - docker tag preprod-sklearn:$FRAMEWORK_VERSION-cpu-py3 515193369038.dkr.ecr.us-west-2.amazonaws.com/sagemaker-scikit-learn:$FRAMEWORK_VERSION-cpu-py3
   post_build:
     commands:

docker/0.23-1/base/Dockerfile.cpu → docker/1.0-1/base/Dockerfile.cpu

@@ -3,8 +3,8 @@ ARG UBUNTU_IMAGE_DIGEST=98706f0f213dbd440021993a82d2f70451a73698315370ae8615cc46
 
 FROM ubuntu:${UBUNTU_VERSION}@sha256:${UBUNTU_IMAGE_DIGEST}
 
-ARG MINICONDA_VERSION=4.8.3
-ARG CONDA_PY_VERSION=37
+ARG MINICONDA_VERSION=4.11.0
+ARG CONDA_PY_VERSION=39
 ARG CONDA_PKG_VERSION=4.9.0
 ARG PYTHON_VERSION=3.7.10
 ARG PYARROW_VERSION=0.16.0
@@ -105,4 +105,4 @@ ENV PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1 PYTHONIOENCODING=UTF-8 LANG=C.U
 # Install Scikit-Learn
 # Scikit-learn 0.20 was the last version to support Python 2.7 and Python 3.4.
 # Scikit-learn now requires Python 3.6 or newer.
-RUN python -m pip install --no-cache -I scikit-learn==0.23.2
+RUN python -m pip install --no-cache -I scikit-learn==1.0.2

docker/1.0-1/base/Dockerfile_aarm64.cpu

@@ -0,0 +1,115 @@
+ARG UBUNTU_VERSION=18.04
+ARG UBUNTU_IMAGE_DIGEST=646942475da61b4ce9cc5b3fadb42642ea90e5d0de46111458e100ff2c7031e6
+
+FROM ubuntu:${UBUNTU_VERSION}@sha256:${UBUNTU_IMAGE_DIGEST}
+
+ARG MINICONDA_VERSION=4.9.2 # Upgraded version
+ARG CONDA_PY_VERSION=39
+ARG CONDA_PKG_VERSION=4.10.1 
+ARG PYTHON_VERSION=3.7.10
+ARG PYARROW_VERSION=1.0
+ARG MLIO_VERSION=arch-agnostic
+
+# Install python and other scikit-learn runtime dependencies
+# Dependency list from http://scikit-learn.org/stable/developers/advanced_installation.html#installing-build-dependencies
+RUN apt-get update && \
+    apt-get -y upgrade && \
+    apt-get -y install --no-install-recommends \
+        build-essential \
+        curl \
+        git \
+        jq \
+        libatlas-base-dev \
+        nginx \
+        openjdk-8-jdk-headless \
+        unzip \
+        wget \
+        && \
+
+    apt-get -y install --no-install-recommends \
+        apt-transport-https \ 
+        ca-certificates \
+	gnupg \
+        software-properties-common \
+        autoconf \
+        automake \
+        build-essential \
+        libssl-dev \
+        && \
+    # MLIO build dependencies
+    # Official Ubuntu APT repositories do not contain an up-to-date version of CMake required to build MLIO.
+    # Kitware contains the latest version of CMake.
+    wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | \
+        gpg --dearmor - | \
+        tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null && \
+    apt-add-repository 'deb https://apt.kitware.com/ubuntu/ bionic main' && \
+    wget https://cmake.org/files/v3.18/cmake-3.18.4.tar.gz && \
+    tar -xzvf cmake-3.18.4.tar.gz && \
+    cd cmake-3.18.4 && \
+    ./configure && \
+    make -j$(nproc) && \
+    make install && \
+    apt-get update && \
+    apt-get install -y --no-install-recommends \
+
+        cmake-data=3.18.4-0kitware1 \
+        doxygen \
+        kitware-archive-keyring \
+        libcurl4-openssl-dev \
+        libtool \
+        ninja-build \
+        python3-dev \
+        python3-distutils \
+        python3-pip \
+        zlib1g-dev \
+        && \
+    rm /etc/apt/trusted.gpg.d/kitware.gpg && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN cd /tmp && \
+    curl -L --output /tmp/Miniconda3.sh https://repo.anaconda.com/miniconda/Miniconda3-py${CONDA_PY_VERSION}_${MINICONDA_VERSION}-Linux-aarch64.sh && \
+    bash /tmp/Miniconda3.sh -bfp /miniconda3 && \
+    rm /tmp/Miniconda3.sh
+
+ENV PATH=/miniconda3/bin:${PATH}
+
+# Install MLIO with Apache Arrow integration
+# We could install mlio-py from conda, but it comes  with extra support such as image reader that increases image size
+# which increases training time. We build from source to minimize the image size.
+RUN echo "conda ${CONDA_PKG_VERSION}" >> /miniconda3/conda-meta/pinned && \
+    # Conda configuration see https://conda.io/projects/conda/en/latest/configuration.html
+   conda config --system --set auto_update_conda false && \
+   conda config --system --set show_channel_urls true && \
+    echo "python ${PYTHON_VERSION}.*" >> /miniconda3/conda-meta/pinned && \
+    conda install -c conda-forge python=${PYTHON_VERSION} && \
+    conda install conda=${CONDA_PKG_VERSION} && \
+    conda update -y conda && \
+    conda install -c conda-forge pyarrow=${PYARROW_VERSION} && \
+    cd /tmp  && \
+    git clone --branch ${MLIO_VERSION} https://github.com/awslabs/ml-io.git mlio && \
+    cd mlio && \
+    build-tools/build-dependency build/third-party all && \
+    mkdir -p build/release && \
+    cd build/release && \
+    cmake -GNinja -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_PREFIX_PATH="$(pwd)/../third-party" ../.. && \
+    cmake --build . && \
+    cmake --build . --target install && \
+    cmake -DMLIO_INCLUDE_PYTHON_EXTENSION=ON -DMLIO_INCLUDE_ARROW_INTEGRATION=ON ../.. && \
+    cmake --build . --target mlio-py && \
+    cmake --build . --target mlio-arrow && \
+    cd ../../src/mlio-py && \
+   python3 setup.py bdist_wheel && \
+    python3 -m pip install --upgrade pip && \
+    python3 -m pip install dist/*.whl && \
+    cp -r /tmp/mlio/build/third-party/lib/libtbb* /usr/local/lib/ && \
+    ldconfig && \
+    rm -rf /tmp/mlio
+
+# Python won’t try to write .pyc or .pyo files on the import of source modules
+# Force stdin, stdout and stderr to be totally unbuffered. Good for logging
+ENV PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1 PYTHONIOENCODING=UTF-8 LANG=C.UTF-8 LC_ALL=C.UTF-8
+
+# Install Scikit-Learn
+# Scikit-learn 0.20 was the last version to support Python 2.7 and Python 3.4.
+# Scikit-learn now requires Python 3.6 or newer.
+ RUN python -m pip install --no-cache -I scikit-learn==1.0.2

docker/0.23-1/extension/Dockerfile.cpu → docker/1.0-1/extension/Dockerfile.cpu

@@ -1,9 +1,9 @@
-FROM preprod-sklearn:0.23-1-cpu-py3
+FROM preprod-sklearn:1.0-1-cpu-py3
 
-RUN pip freeze | grep -q 'scikit-learn==0.23.2'; \
+RUN pip freeze | grep -q 'scikit-learn==1.0.2'; \
 	if [ $? -eq 0 ]; \
-		then echo 'scikit-learn version 0.23.2 requirement met'; \
-		else echo 'ERROR: Expected scikit-learn version is 0.23.2, check base images for scikit-learn version' && \
+		then echo 'scikit-learn version 1.0.2 requirement met'; \
+		else echo 'ERROR: Expected scikit-learn version is 1.0.2, check base images for scikit-learn version' && \
 			 exit 1; fi
 
 RUN pip install --upgrade --no-cache --no-deps sagemaker-scikit-learn-extension==2.5.0

docker/0.23-1/extension/README.md → docker/1.0-1/extension/README.md

@@ -8,15 +8,15 @@ The SageMaker Scikit-learn Extension Container is built in 3 steps. The first 2
 
 The "base" Dockerfile encompass the installation of the framework and all of the dependencies needed.
 
-Tagging scheme is based on <Scikit-learn_version>-<SageMaker_version>-cpu-py<python_version>. (e.g. 0.23-1-cpu-py3)
+Tagging scheme is based on <Scikit-learn_version>-<SageMaker_version>-cpu-py<python_version>. (e.g. 1.0-1-cpu-py3)
 
 All "final" Dockerfiles build images using base images that use the tagging scheme above.
 
 ```
-docker build -t sklearn-base:0.23-1-cpu-py3 -f docker/0.23-1/base/Dockerfile.cpu .
+docker build -t sklearn-base:1.0-1-cpu-py3 -f docker/1.0-1/base/Dockerfile.cpu .
 ```
 
-Notice that this Dockerfile has the updated version of sklearn (0.23.2) installed.
+Notice that this Dockerfile has the updated version of sklearn (1.0.2) installed.
 
 ### Step 2: Final Image
 
@@ -38,7 +38,7 @@ python setup.py bdist_wheel
 Then build the final image, like in the sagemaker-sklearn-container
 
 ```
-docker build -t preprod-sklearn:0.23-1-cpu-py3 -f docker/0.23-1/final/Dockerfile.cpu .
+docker build -t preprod-sklearn:1.0-1-cpu-py3 -f docker/1.0-1/final/Dockerfile.cpu .
 ```
 
 ### Step 3: Build the extension image for SageMaker Scikit-learn Extension Container
@@ -47,10 +47,10 @@ The "extension" Dockerfiles encompass the installation of the SageMaker Autopilo
 
 The "extension" Dockerfiles use final images for building.
 
-Build the third additional Dockerfile needed for SageMaker Scikit-learn Extension Container. This Dockerfile specifies a hard dependency on a certain version of scikit-learn (i.e. v0.23.2).
+Build the third additional Dockerfile needed for SageMaker Scikit-learn Extension Container. This Dockerfile specifies a hard dependency on a certain version of scikit-learn (i.e. v1.0.2).
 
 Tagging scheme is based on extension-<Scikit-learn-Extension_version>-<SageMaker_version>-cpu-py<python_version>. (e.g. extension-2.5-1-cpu-py3). Make sure the "extension" image is tagged in accordance with the  `extension` (i.e. `extension-2.5-1-cpu-py3`).
 
 ```
-docker build -t preprod-sklearn-extension:2.5-1-cpu-py3 -f  docker/0.23-1/extension/Dockerfile.cpu .
+docker build -t preprod-sklearn-extension:2.5-1-cpu-py3 -f  docker/1.0-1/extension/Dockerfile.cpu .
 ```

docker/0.23-1/final/Dockerfile.cpu → docker/1.0-1/final/Dockerfile.cpu

@@ -1,5 +1,5 @@
-FROM sklearn-base:0.23-1-cpu-py3
-ENV SAGEMAKER_SKLEARN_VERSION 0.23-1
+FROM sklearn-base:1.0-1-cpu-py3
+ENV SAGEMAKER_SKLEARN_VERSION 1.0-1
 
 LABEL com.amazonaws.sagemaker.capabilities.accept-bind-to-port=true
 
@@ -54,4 +54,3 @@ ENV TEMP=/home/model-server/tmp
 
 # Required label for multi-model loading
 LABEL com.amazonaws.sagemaker.capabilities.multi-models=true
-

docker/1.0-1/final/Dockerfile_aarm64.cpu

@@ -0,0 +1,56 @@
+FROM sklearn-base:1.0-1-arm64-cpu-py3
+ENV SAGEMAKER_SKLEARN_VERSION 1.0-1
+
+LABEL com.amazonaws.sagemaker.capabilities.accept-bind-to-port=true
+
+COPY requirements.txt /requirements.txt
+RUN python -m pip install -r /requirements.txt && \
+    rm /requirements.txt
+
+COPY dist/sagemaker_sklearn_container-2.0-py3-none-any.whl /sagemaker_sklearn_container-2.0-py3-none-any.whl
+# # https://github.com/googleapis/google-cloud-python/issues/6647
+RUN rm -rf /miniconda3/lib/python3.7/site-packages/numpy-1.19.4.dist-info && \
+    pip install --no-cache /sagemaker_sklearn_container-2.0-py3-none-any.whl && \
+    rm /sagemaker_sklearn_container-2.0-py3-none-any.whl
+
+ENV SAGEMAKER_TRAINING_MODULE sagemaker_sklearn_container.training:main
+ENV SAGEMAKER_SERVING_MODULE sagemaker_sklearn_container.serving:main
+
+#######
+# MMS #
+#######
+# Create MMS user directory
+RUN useradd -m model-server
+RUN mkdir -p /home/model-server/tmp
+RUN chown -R model-server /home/model-server
+
+# Copy MMS configs
+COPY docker/$SAGEMAKER_SKLEARN_VERSION/resources/mms/config.properties.tmp /home/model-server
+ENV SKLEARN_MMS_CONFIG=/home/model-server/config.properties
+
+# Copy execution parameters endpoint plugin for MMS
+RUN mkdir -p /tmp/plugins
+COPY docker/$SAGEMAKER_SKLEARN_VERSION/resources/mms/endpoints-1.0.jar /tmp/plugins
+RUN chmod +x /tmp/plugins/endpoints-1.0.jar
+
+# Create directory for models
+RUN mkdir -p /opt/ml/models
+RUN chmod +rwx /opt/ml/models
+
+#####################
+# Required ENV vars #
+#####################
+# Set SageMaker training environment variables
+ENV SM_INPUT /opt/ml/input
+ENV SM_INPUT_TRAINING_CONFIG_FILE $SM_INPUT/config/hyperparameters.json
+ENV SM_INPUT_DATA_CONFIG_FILE $SM_INPUT/config/inputdataconfig.json
+ENV SM_CHECKPOINT_CONFIG_FILE $SM_INPUT/config/checkpointconfig.json
+
+# Set SageMaker serving environment variables
+ENV SM_MODEL_DIR /opt/ml/model
+
+#EXPOSE 8080
+ENV TEMP=/home/model-server/tmp
+
+# Required label for multi-model loading
+LABEL com.amazonaws.sagemaker.capabilities.multi-models=true

requirements.txt

@@ -15,6 +15,9 @@ retrying==1.3.3
 sagemaker-containers==2.8.6.post2
 sagemaker-inference==1.2.0
 sagemaker-training==4.0.1
-scikit-learn==0.23.2
+scikit-learn==1.0.2
 scipy==1.5.3
 six==1.15.0
+jinja2==2.10.2
+MarkupSafe==1.1.1
+Werkzeug==0.15.6

setup.py

@@ -32,6 +32,7 @@ def read(fname):
         "Programming Language :: Python",
         "Programming Language :: Python :: 3.6",
         "Programming Language :: Python :: 3.7",
+        "Programming Language :: Python :: 3.9",
     ],
 
     install_requires=read("requirements.txt"),

test/conftest.py

@@ -43,7 +43,7 @@ def pytest_addoption(parser):
     parser.addoption('--install-container-support', '-C', action='store_true')
     parser.addoption('--docker-base-name', default='sk-learn')
     parser.addoption('--region', default='us-west-2')
-    parser.addoption('--framework-version', default='0.20.0')
+    parser.addoption('--framework-version', default='1.0.2')
     parser.addoption('--py-version', choices=['2', '3'], default=str(sys.version_info.major))
     parser.addoption('--processor', choices=['cpu'], default='cpu')
     # If not specified, will default to {framework-version}-{processor}-py{py-version}

test/integration/test_multiple_model_endpoint.py

@@ -56,18 +56,22 @@ def modulevolume():
 @pytest.fixture(scope='module', autouse=True)
 def container(request, docker_base_name, tag):
     test_name = 'sagemaker-sklearn-serving-test'
+    module_dir = os.path.join(resource_path, 'module')
+    model_dir = os.path.join(resource_path, 'models')
     try:
         command = (
             'docker run --name {} -p 8080:8080'
-            ' --mount type=volume,source=dynamic_endpoint_model_volume,target=/opt/ml/model,readonly'
-            ' --mount type=volume,source=dynamic_endpoint_module_volume,target=/user_module,readonly'
+            # ' --mount type=volume,source=dynamic_endpoint_model_volume,target=/opt/ml/model,readonly'
+            # ' --mount type=volume,source=dynamic_endpoint_module_volume,target=/user_module,readonly'
+            ' -v {}:/opt/ml/model'
+            ' -v {}:/user_module'
             ' -e SAGEMAKER_BIND_TO_PORT=8080'
             ' -e SAGEMAKER_SAFE_PORT_RANGE=9000-9999'
             ' -e SAGEMAKER_MULTI_MODEL=true'
             ' -e SAGEMAKER_PROGRAM={}'
             ' -e SAGEMAKER_SUBMIT_DIRECTORY={}'
             ' {}:{} serve'
-        ).format(test_name, 'script.py', "/user_module/user_code.tar.gz", docker_base_name, tag)
+        ).format(test_name, model_dir, module_dir, 'script.py', "/user_module/user_code.tar.gz", docker_base_name, tag)
 
         proc = subprocess.Popen(command.split(), stdout=sys.stdout, stderr=subprocess.STDOUT)